A Clustering Approach to a Major-Accident Data Set: Analysis of Key Interactions to Minimise Human Errors

This work aims to scrutinise a proprietary dataset containing major accidents occurred in high-technology facilities, in order to disclose relevant features and indicate a path to the recognition of the genesis of human errors. The application of a tailored Hierarchical Agglomerative Clustering method, using the bray-curtis dissimilarity and two different linkage functions - complete and average - will provide means to understand data and identify key similarities among accidents. Significant interfaces between human factors, the organisational environment and the technology will be described. Main clustering results have shown that accidents featuring communication issues and interface problems were grouped together, and turned out to be the most deadly ones, considering the fatality rate. Another cluster highlighted relevant training shortcomings. Also, design failures, poor quality control and inadequate task allocation were identified as key contributing factors to major accidents. Conclusions to improve the human performance based on these clustering results are then discussed.